An ensemble classifier for vibration-based quality monitoring

Vibration-based quality monitoring of manufactured components often employs pattern recognition methods. Albeit developing several classification methods, they usually provide high accuracy for specific types datasets, but not general cases. In this paper, issue has been addressed by a novel ensemble classifier based on the Dempster-Shafer theory evidence. proposed procedure, prior to DST combination, three steps should be taken: (i) selection proper classifiers maximizing joint mutual information between predicted and target outputs, (ii) optimal redistribution classifiers’ outputs considering distance (iii) utilizing five different weighting factors enhance fusion performance. The effectiveness framework is validated its application 13 UCI KEEL machine learning datasets. It then applied two vibration-based datasets detect defected samples: one synthetic dataset generated from finite element model dogbone cylinder, real experimental collecting broadband vibrational response polycrystalline Nickel alloy first-stage turbine blades. investigation made through statistical analysis in presence noise levels. Comparing results with those state-of-the-art techniques reveals good performance method.